1
Biorhythms of FX:
FX Modeling & Risk-Return Assessment
By: Jeremy Fand, Global Head of FX Strategy @UBS Warburg (203) 719-8160
Overview : A project focusing on creating quantitative risk-return assessment processes to aid in making prudent currency decisions and properly structuring currency portfolios and currency overlay portfolios, including the use of spot, forward, and options structures.
Advising the customer in navigating the choppy FX market has matured dramatically in recent years. While daily and weekly publications that focus on the news, views, and tech levels driving markets are the mainstay of FX strategy desks, and while these are clearly essential items, the client has demanded more. Indeed, the market not only wants the latest read on the fundamental and technical factors that are driving the markets, but now demand methodologies that help them manage these exposures in a way that balances risk and return expectations across their whole portfolio.
Going beyond the strategist’s interpretation of market events will allow UBS Warburg to deepen client relationships and obtain greater order flow through the provision of additional value-added services built around a disciplined and rigorous modeling approach. The models we are developing can help discern and then quantify the strength of the themes that are driving markets. In a sense, the models become the ‘biorhythm of FX’ that dynamically identify the next theme that comes into vogue. Technically, this is accomplished by using dynamic statistical process that constant update factor “betas” (estimated coefficients) for those economic and financial variables that represent the themes on which our strategy team is focusing. We can also shed quantitative light on such perennial questions as whether an exchange rate is moving because (a) one currency is weak, (b) the other currency is strong, or (c) some external event has shifted the relative terms of trade and capital between the two currencies.
Understanding Bayesian Statistical Analysis
The Reverend Thomas Bayes, living in the 1700s in England, pioneered a brand of statistical analysis that involves the continuous updating of the expected probability distribution of future events occurring based on any new information, data, or theoretical insights that may come available since the last time the expected probability distribution was revised. This Bayesian statistical framework is inherently dynamic. It allows one to track in a quantitative manner how estimated probability distributions evolve and which factors are contributing to the change in the new, and updated estimated probability distribution.
Traditional statistical studies tend to pick a fixed period of time, and then conduct a standard set of statistical regressions to determine the impact of specific factors or explanatory variables for whole period. In this traditional approach, the impact of a specific factor (i.e., its “Beta” coefficient) is fixed for the whole period under analysis. A shift in regime is a problem for the statistical “frequentist.” Usually they would have to divide their study into two separate time periods and assume a fixed, but possibly different, contribution for each factor when comparing the two different periods. By contract, a Bayesian approach works dynamically to allow the contribution of certain factors, as measured by their “Beta” coefficients to evolve over time. One can even study the pace at which the changes take place, since not all things change instantly, or at the same time, or in a well-defined step pattern, as the “Frequentist” approach assumes. Thus, for the study of financial markets, a Bayesian approach potentially can be much more useful than the competing “Frequentist” approach.
Factor analysis over traditional econometrics
Fundamental currency analysis is an imperfect science. Themes that drive markets wax and wane through time, and do not always adhere to the economic textbook. Traditional fundamental analysis assumes that certain economic rules translate into defined currency movements. In practice, however, that is not always true. For example, the economics textbook would tell you that interest rate differentials will drive currencies, yet often interest rate differentials are inversely correlated to the currency movement. Indeed, typical fundamentals-based econometric models of exchange rates tend to work well for select periods and select currencies. Different periods and currencies, however, require different models – some even counter to textbook economics. The challenge is to identify which model(s) will best fit a particular currency for the upcoming period. In other words, factor analysis is key to explaining the FX markets dynamic movements and theme accountability.
Factor analysis needs better statistical discipline
The calls generated by different econometric sub-models often conflict. For example, an interest rate differential model may generate a call to sell yen while the trade balance model is generating a call to buy yen. Resolving this conflict requires better statistical tools. The typical result of such a conflict is that the weighted average call will be neutral, unless one model is given much more credence than the other. Bayesian statistical methods can help generate weightings for the different factors driving markets and generate a single directional call. Because there are many potentially conflicting economic forces that drive exchange rates, the FX Strategy Team writes daily research publications that are designed to provide an analysis of the tug-of-war between these forces. If our strategy team can see that one side of the tug of war is clearly stronger, we make an investment decision and notify our clients. The goal of this project is to build a set of quantitative tools that analyze exchange rates exactly the way we do. As a quantitative model, the project is a systematic compliment to the subjective analysis of the strategists. While the model uses sophisticated mathematics to calculate the strength on each end of the rope in the exchange rate tug-of-war, it is the human strategists who analyze the exchange rate tug-of-war in real time by sight and can focus more on information that doesn't come in the form of a number.
Quantitative Tools for FX Research
The first stage of the project is integrating single-equation FX factor models into a multi-equation simultaneous and dynamic Bayesian estimation system. The models estimate risk and correlation information along with return forecasts that can be interpreted in the light of the factors responsible for the estimates. The product thus enables us to discuss the ‘betas’ or strength of themes underlying currency movements, and map out what themes are driving markets. More importantly, we can now discuss views on currency markets in a portfolio context accompanied by a sophisticated risk management discipline. The second stage of the project is to adapt these models with subjective input – i.e., expert information. Recognizing that the client’s views regarding fundamental forces may not always correspond with ours, the models will allow for expert information entry and scenario analysis. This stage of the process will allow for scenario analysis, style risk analysis, and user specific forecasting of both spot and volatility.
Risk – Return Analysis
Because of the way in which the FX Biorhythms product focuses on constructing risk-return efficient currency portfolios, we can work together with the client in understanding not only the drivers of FX movements, but how best to express views. We can therefore serve not only those customers with a “currency picking” focus but also serve those with a “currency portfolio management” focus, with appropriate attention to risks and diversification (correlations) as well as innovative swaps and options structures to fit asymmetric portfolio risk-return requirements. Our modeling process connects directly to our quantitative advisory team as the risk assessment process lends itself naturally to distinguishing different volatility characteristics and developing options structures to meet those needs.
Integration of Technicals and Fundamental Analysis
The dynamic factor analysis with a quantitative risk-return assessment process could also be used to integrate fundamental, technical and flow analysis. Input into the modeling process can be market information, economic data or technical regimes. Bayesian statistical methods gives a good framework for integrating disparate factors into one clean system. Traditionally, technical analysis and fundamental analysis have been separate pillars of FX prediction. Integrating the two has usually meant overweighing the fundamentals during periods of risk aversion (credit modeling) and overweighing technicals in risk loving times (momentum modeling). A system that integrates factors from both – even on different time frames, would be a unique contribution. With the growing attention on flow information, we include our own proprietary data as a third regime.